A comprehensive review of recommender systems: Transitioning from theory to practice — *Computer Science Review*, February 2026.
- Source: https://www.sciencedirect.com/science/article/pii/S157401372500125X
- Type: Survey / Review
- Recommender systems have long faced challenges around efficiency, scalability, and user-intent understanding; traditional pipelines separate retrieval, ranking, and generation, which is ill-suited to LLM-era needs for natural-language interaction, multi-hop reasoning, and up-to-date knowledge.
- The survey provides a unified framework covering:
- Modeling paradigms: discriminative vs. generative retrieval (two-tower models, cross-encoders, DSI, document-id generation).
- LLM integration: RAG, search agents, and tool/API calling — flexible and interpretable, but with latency and error-propagation costs.
- Optimization objectives: relevance, diversity, and freshness via multi-objective LTR, RLHF, and online learning.
- Evaluation: offline (nDCG, MRR, Recall@k), online (A/B testing), and human/LLM-as-judge protocols.
- Historical timeline: 2019–2021 — BERT reranking and DPR establish neural retrieval; 2022–2023 — RAG and FreshLLM drive retrieval-generation fusion; 2024 onward — conversational/agentic search and Gen-RecSys surge; 2025–2026 — RL-trained search agents, Deep Research, and GraphRAG emerge as growth areas.
- Open problems: lack of unified benchmarks, irreproducible private data, LLM evaluation bias, safety and cost constraints for agentic systems, cross-lingual and multimodal generalization.
- A Comprehensive Survey on Retrieval Methods in Recommender Systems (arXiv:2407.21022)
- A Review of Modern Recommender Systems Using Generative Models (Gen-RecSys), DOI: 10.1145/3637528.3671474
- A Survey of Generative Search and Recommendation in the Era of Large Language Models (arXiv:2404.16924)
- A Survey on LLM-powered Agents for Recommender Systems (arXiv:2502.10050)
- A Comprehensive Survey on Cross-Domain Recommendation (arXiv:2503.14110)
Key points
Practical implications
1. Architecture: cascaded retrieval + reranking + generation remains mainstream, but agentic paradigms make retrieval strategy itself a learnable object. 2. Data: high-quality instruction data and click/session logs matter; synthetic data requires care to avoid leakage and distribution shift. 3. Evaluation: offline/online gaps are widening; LLM-as-judge should be cross-validated with human assessment. 4. Deployment: latency, cost, explainability, and safety are hard industrial constraints — optimizing academic benchmarks alone is insufficient.
> Note: This entry is compiled from the paper's abstract and public metadata. Refer to the original PDF for exact tables and quantitative results.